Embedded Files
Anak Agung Krisna Ananda Kusuma, Jung-Su Kim*
Seoul National University of Science and Technology
anak_agung86@seoultech.ac.kr, jungsu@seoultech.ac.kr
Abstract
Abstract
Classical map-based approaches to autonomous racing are known for their high performance, as they can compute globally optimal reference trajectories for the vehicle to follow. However, these methods are limited to pre-mapped environments, which restricts their applicability in fully autonomous systems that must operate in inherently unknown or unmapped settings. In contrast, mapless racing approaches eliminate the need for a global map by relying solely on sensor data to construct a local representation of the environment. Despite their advantage of not requiring global maps, existing mapless methods often fall short in performance, particularly struggling to sustain high speeds through corners. To overcome this limitation, this paper propose a receding horizon minimum-curvature based trajectory planner that operates within local map framework, enabling high performance racing using only local observations and removing the reliance on global reference. The proposed method is integrated with a Model Predictive Control (MPC) scheme, which ensures accurate trajectory tracking while respecting the vehicle’s physical constraints. As a result, the vehicle is able to take corners more effectively, advancing the capabilities of autonomous racing in unknown environments. Both Gazebo simulations and real-world experiments demonstrate that the proposed mapless racing method achieves better performance compared to the other two mapless methods.
Motivation: Enhancing autonomous mapless racing performance by improving corner handling
Current mapless approaches often struggle to deliver strong performance, primarily due to ineffective corner handling. This results in sharper turns, reduced speed, and consequently longer lap times.
Proposed method: Minimum curvature-based receding horizon trajectory generation and tracking in autonomous mapless racing.
A minimum-curvature-based trajectory generation and tracking method that leverages odometry-based transformation is proposed. This approach allows for trajectory tracking scheme in local map framework, allowing the vehicle to navigate corners more effectively.
Simulation
Simulation were conducted using a high-fidelity Gazebo simulation. The simulation employs an Ackermann-type vehicle equipped with a LiDAR sensor for trajectory generation, as well as an IMU and wheel encoders integrated with a Kalman filter for odometry estimation.
System architecture
Comparison
Two other mapless racing methods, namely the Disparity Extender (Baseline 1) and the Two-Stage (Baseline 2) method, were used for performance comparison with the proposed method. The evaluation was conducted on two racetracks, Mco and Aut. Each method was run for six laps, and the true states were logged from the Gazebo simulation to compute the mean, maximum, minimum, and variance of the raceline curvature and velocity profiles along the arc length of the resulting raceline.
Raceline
The resulting raceline for global pre-mapped, Disparity Extender, Two-Stage, and proposed method in Mco racetrack.
The resulting raceline for global pre-mapped, Disparity Extender, Two-Stage, and proposed method in Aut racetrack.
Curvature
Simulation curvature comparison: The raceline curvature produced by the proposed method is most similar to the pre-mapped global plan compared to the other methods.
Velocity
Simulation velocity comparison: The proposed method achieves higher average speeds through corners.
Simulation lap time result
lap times were recorded by detecting the vehicle’s ground-truth position in the simulation as it passed a circle centered at (0,0) with a radius of 0.5 m. The result shows that the proposed method achieves the lowest lap time.
Simulation video
The videos below showcase each of the three methods in action on the Mco and Aut racetracks.
Explanation video of the proposed method
Proposed method Mco racetrack
Two-Stage method Mco racetrack
Disparity Extender method Mco racetrack
Proposed method Aut racetrack
Two-Stage method Aut racetrack
Disparity Extender method Aut racetrack
Real-world Experiment
Experiments were conducted using the F1/10 autonomous racing platform. The vehicle is a 1:10 scale model of a full-sized car and is equipped with a Hokuyo LiDAR sensor, a VESC electronic speed controller with encoder, and an MW-AHRSv2U IMU sensor. The main onboard computing unit used is the NVIDIA Jetson Orin NX.
The experiments were conducted in a simplified indoor arena with sharp corners, serving to demonstrate the real-world applicability of the proposed method.
Comparison
Two other mapless racing methods, namely the Disparity Extender (Baseline 1) and the Two-Stage (Baseline 2) method, were used for performance comparison with the proposed method. Each method was run for six laps, with the vehicle’s pose and velocity recorded in the global frame using the gmcl localization technique, and subsequently processed to obtain curvature and velocity profiles along the arc length.
Raceline
The resulting raceline for global pre-mapped, Disparity Extender, Two-Stage, and proposed method in real-world experiment.
Curvature
Real-world experiment curvature comparison: The raceline curvature produced by the proposed method is most similar to the pre-mapped global plan compared to the other methods.
Velocity
Real-world experiment velocity comparison: The proposed method achieves higher average speeds through corners.
Simulation lap time result
lap times were recorded by detecting the vehicle’s pose in the global frame as it passed a circle centered at (0,0) with a radius of 0.5 m. The result shows that the proposed method achieves the lowest averaged lap time.
Experiment video
The videos below showcase each of the three methods in action for the real-world experiment
Proposed method real-world experiment
Two-Stage method real-world experiment
Disparity Extender method real-world experiment
For more videos about the project, please visit www.youtube.com/@cdslseoultech4967
Page updated
Google Sites
Report abuse